JPH04261591A - Automatic music scoreing device - Google Patents

Abstract

PURPOSE:To satisfactorily extract the basic frequency by extracting a basic frequency candidate from a power spectrum obtained by executing a frequency analysis at every prescribed time with respect to data. CONSTITUTION:A signal fetching part 21 executes an A/D conversion of music, and sets it to a digital signal handled in a computer. Also, a frequency analysis processing part 22 calculates a power spectrum in the frequency direction within a prescribed time by executing a frequency analysis at every certain prescribed time with respected to data fetched by the signal fetching part 21. Subsequently, a basic frequency candidate extracting part 23 extracts a basic frequency from the power spectrum.

Description

[Detailed description of the invention]

0001

[Field of Industrial Application] The present invention relates to an automatic score transcription device that converts music into a musical score or a code corresponding to a musical score, and more specifically, the present invention relates to an automatic score transcription device that converts music into a musical score or a code corresponding to a musical score. Regarding.

0002

2. Description of the Related Art Conventionally, music played by a plurality of musical instruments has been transcribed by a transcriber with musical knowledge. In addition, in the case where music is automatically transcribed by a device rather than by a person, there have been proposed devices that transcribe music consisting of single notes or that transcribe music based on information about keyboard presses.

0003

[Problems to be Solved by the Invention] However, with conventional music transcription devices, there are restrictions on the types and number of instruments, which limits the range of music transcription, and there is a problem in that it is not possible to transcribe general musical performances as they are. . Another problem is that if you simply frequency-analyze the A/D-converted signal of music played by multiple instruments using FFT, etc., many harmonics of each instrument will be observed, making it impossible to easily determine which sound is the fundamental frequency. was there. The present invention solves the above-mentioned problems by frequency-analyzing the A/D-converted music signal, removing overtones from the result, and extracting only the fundamental frequency. The purpose is to provide an automatic score transcription device that is not subject to restrictions.

0004

Means for Solving the Problems In order to achieve the above object, the invention of claim 1 provides a score recording device that converts a music signal into a musical score or a code corresponding to the musical score, in which a music signal is taken in and A/D converted. an acquisition unit; a frequency analysis processing unit that calculates a power spectrum in the frequency direction within a certain time by performing frequency analysis on the data acquired by the signal acquisition unit at certain time intervals; and a fundamental frequency candidate extraction section that extracts fundamental frequency candidates from the spectrum. The invention according to claim 2 is characterized in that the fundamental frequency candidate extracting unit includes power threshold calculation means for calculating a power threshold from the peak of the power spectrum, and Determines whether a performance note has the fundamental frequency based on the frequency and power of the peak that is a harmonic of the original peak among other peaks with higher frequencies, the original peak, and the peaks between the two. fundamental frequency retrieval means for determining whether one of the peaks of the power spectrum is a harmonic of any of the other peaks having a lower frequency; a harmonic series content calculation means for calculating to what extent the harmonic series of the peak searched by the frequency search means is included in the peaks of the power spectrum; The overtone removing means removes the peak determined to be an overtone from the power spectrum when the peak is determined to be an overtone of another peak.

[0005]

[Operation] According to the structure of claim 1, the signal acquisition section:
Converts music from A/D to a digital signal that can be handled within a computer. The frequency analysis processing section performs frequency analysis on the data acquired by the signal acquisition section at certain fixed time intervals, thereby calculating the power/power in the frequency direction within the fixed time period.
Calculate the spectrum. The fundamental frequency candidate extraction section extracts fundamental frequency candidates from this power spectrum. Each means constituting the fundamental frequency candidate extracting section described in claim 2 operates as follows. The power threshold calculation means calculates a power threshold from the peak of the power spectrum. The fundamental frequency similarity determining means determines whether one of the peaks in the power spectrum is a fundamental frequency or not, and compares it with a peak that is a harmonic of the original peak among other peaks with higher frequencies, and a peak that is a harmonic of the original peak. , based on the frequency and power of the peak between the two. The fundamental frequency search means searches whether one of the peaks of the power spectrum is a harmonic of any of the other peaks having a lower frequency. The harmonic series content calculating means calculates to what extent the harmonic series of the peak searched by the fundamental frequency searching means is included in the peaks of the power spectrum. The overtone removing means removes the peak determined to be an overtone from the power spectrum when it is determined that one of the peaks of the power spectrum is an overtone of another peak.

[0006]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment embodying the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram of an automatic score transcription apparatus according to the present invention. This device includes an audio amplifier 1 to which a music signal is input, a low-pass filter 2, and an A
/D conversion device 3, I/O port 4, CPU 5 for extracting fundamental frequency candidates, RAM 6, ROM 7,
It is composed of a display 8 that displays the extracted results.

Next, by adding FIGS. 2 and 3 to FIG. 1 above, the functional configuration for fundamental frequency extraction processing performed by the automatic score transcription apparatus will be explained. This device includes a signal acquisition section 21, a fast Fourier transform (FFT) processing section 22, and a fundamental frequency candidate extraction section 23 as functional components. The signal acquisition unit 21 includes an audio amplifier 1, a low-pass
Filter 2, A/D converter 3, I/O port 4, CP
Consists of U5 etc. The input music signal is amplified by the audio amplifier 1, and this amplified signal is input to the low-pass filter 2, for example, 5.5k.
Only frequency components below Hz pass through, suppressing aliasing distortion during sampling. This output signal is transmitted to the A/D converter 3
The signal is sampled at 12 kHz and 16 bits. The sampled data is taken into the CPU 5 via the I/O port 4 and stored in the RAM 6.

[0008] The FFT processing section 22 includes a CPU 5 and a RAM 6.
etc., the CPU 5 reads the sampled data from the RAM 6, takes every 25 msec as one frame, multiplies an 85.3 msec Hamming window for each frame, and then reads the sampled data from the FF
A logarithmic power spectrum is calculated by T analysis. Next, the CPU 5 determines the peak frequency from the calculated logarithmic power spectrum by radiation interpolation processing. FIG. 5 shows the peak spectrum obtained as described above, with the time axis on the horizontal axis and the peak frequency converted into a keyboard number on the vertical axis, and the intensity shown in shading. The figure shows data before overtone removal. The fundamental frequency candidate extracting unit 23 uses the CPU
5, and the power threshold calculation means 31 as shown in FIG.
, a fundamental frequency likelihood determining means 32 , a fundamental frequency searching means 33 , an overtone series content calculating means 34 , and an overtone removing means 35 .

[0009] The fundamental frequency candidate extracting section 23 advances the processing for each frame of the FFT processing section 22. Now, it is assumed that there are N peaks in the spectrum obtained by the FFT processing unit 22 for a certain frame, and the following description will be made according to the flowchart of FIG. FIG. 4 shows the fundamental frequency candidate extraction unit 23
This figure shows the processing for one frame. Regarding the N peaks, the frequency F and power value P are sorted by frequency and stored in the RAM 6. A power threshold Sp is calculated from the power values of these N peaks (step #1). Regarding the N peaks, processing is performed in order from the highest frequency to the lowest frequency. Of N peaks, n from the highest frequency
Let the frequency of the th peak be F(n) and the power be P(n). Check whether F(n) is the fundamental frequency by F(n
) higher frequency F(m) and power P(m)(1
Judgment is made from ≦m<n) (#2, #3). This determination is made as follows. First, a search is made to see if there is an F(x) in F(m) that is the X-th harmonic (X≧2) of F(n). If F(x) is found, then F(n) and F(
x) relationship is when F(n) is the fundamental frequency and F(x) is the There may be a case (this is referred to as case 2) where the low frequency is an L-order (L≧2) or L×X-order overtone of a frequency that is a common divisor of both.

In order to determine which of these cases is the case, the frequency F(y) that exists between F(x) and F(n) is
, and power P(y) (x<y<n), the following processing is performed. P(y)>(P(x)+P(n)
)/2, calculate the frequency difference between F(y), F(x), and F(n). Now, Fxy=F(x)-F(y),
Let Fyn=F(y)−F(n). The smaller of Fxy and Fyn is set as F0. If F0 has both F(n) and F(x) as overtones, the above-mentioned case 2 is likely, and therefore F(n) is determined not to be the fundamental frequency. Also, if F(x) is not found, it is determined that F(n) is not the fundamental frequency. In other cases, that is, the Xth harmonic of F(n) existing in F(m) (X≧2
) for all F(x), F(x) and F
All F calculated from F(y) existing between (n)
If there is no one in which both F(n) and F(x) are overtones, F(n) is determined to be the fundamental frequency.

From the above, P(n)>Sp and F(n)
If it is determined that is the fundamental frequency (YES in #4, #5), that peak is determined to be the fundamental frequency, and processing is shifted to the next peak (#6, #7). If not,
Search for F(z) with F(n) as the Z-th harmonic (Z≧2) among the frequencies F(p) (n<p≦N) lower than F(n) (#8 , #9). If F(z) is found, the overtone series up to the Zth order of F(z) is expressed as F(q)(z≦
Search within q≦n). Of the Z overtone series of F(z) that should be included in F(q), let W be the number actually found in F(q), and calculate the overtone series content W/Z. (#10). If W/Z is above a certain threshold (YES in #11), F(n) is considered to be an overtone of F(z) and is removed (#12). If F(z) that satisfies the conditions is not found, F(n) is not removed. Note that even if F(n) is removed, the information that a peak exists in F(n) is left so as not to interfere with subsequent processing. After the above processing is performed for all N peaks, processing is moved to the next frame. When the FFT processing section 22 completes the above processing for all frames for which power spectra have been calculated, the processing of the fundamental frequency candidate extraction section 23 ends.

FIG. 6 shows the data after the data shown in FIG. 5 has been subjected to overtone removal processing by the fundamental frequency candidate extraction section 23. In FIG. 6, compared to FIG. 5, it can be seen that unnecessary data has been removed by overtone removal. The present invention is not limited to the above embodiments, and various changes can be made without departing from the spirit thereof. for example,
In this embodiment, the fundamental frequency candidate extracting section 23 processes each frame of the FFT processing section 22, but the section in which the stable part of the power spectrum continues between adjacent frames is set in advance as one analysis section. , it is also possible to perform subsequent processing for each analysis section.

[0013]

As described above, according to the present invention, by frequency-analyzing the A/D-converted music signal and removing overtones from the result, only the fundamental frequency of the sound actually being played can be obtained. , the fundamental frequency can be extracted without being restricted by the type of instrument or the number of instruments, making it easy to transcribe.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of an automatic score transcription apparatus according to an embodiment of the present invention.

FIG. 2 is a functional configuration diagram of the automatic score transcription device up to fundamental frequency extraction.

FIG. 3 is a configuration diagram of a fundamental frequency candidate extraction unit 23.

FIG. 4 is a flowchart showing processing in the fundamental frequency candidate extracting unit 23.

FIG. 5 is an explanatory diagram of a power spectrum calculated by the FFT processing unit 22 before overtone removal processing.

6 is an explanatory diagram of the power spectrum after the overtone removal process is performed by the fundamental frequency candidate extraction unit 2 in FIG. 5. FIG.

[Explanation of symbols]

3 A/D conversion device 5 CPU 6 RAM 21 Signal acquisition section 22 FFT processing section 23 Fundamental frequency candidate extraction unit 31 Power threshold calculation means 32 Fundamental frequency likeness determination means 33 Fundamental frequency search means 34 Overtone series content calculation means 35 Overtone removal means

Claims (2)

[Claims] Claim 1. A score transcription device that converts a music signal into a musical score or a code corresponding to the musical score, which includes a signal acquisition section that takes in the music signal and performs A/D conversion, and a signal acquisition section that converts the music signal into an A/D converter, and a signal acquisition section that converts the music signal into a musical score or a code corresponding to the musical score. a frequency analysis processing unit that calculates a power spectrum in the frequency direction within a certain period of time by performing frequency analysis at each time; and a fundamental frequency candidate extraction unit that extracts fundamental frequency candidates from the power spectrum. An automatic music transcription device featuring: 2. The fundamental frequency candidate extraction section includes a power threshold calculation means for calculating a power threshold from a peak of a power spectrum, and one of the peaks of the power spectrum is a fundamental frequency candidate of a performance sound. The fundamental frequency, which is determined based on the frequency and power of the peak that is a harmonic of the original peak among other peaks at higher frequencies, the original peak, and the peaks between them. similarity determining means, fundamental frequency searching means for searching whether one of the peaks of the power spectrum is a harmonic of any of the other peaks having a lower frequency, and the fundamental frequency searching means a harmonic series content calculating means for calculating to what extent the harmonic series of the searched peak is included in the peaks of the power spectrum; 2. The automatic score transcription apparatus according to claim 1, further comprising an overtone removing means for removing the peak determined to be an overtone from the power spectrum when the peak is determined to be an overtone.